Translated AI Approaches Human Parity in Translation Editing Speed by Decade End
AI translation company Translated observed its AI's editing time drop to 2 seconds per word by 2022, suggesting human-level editing effort could be reached by the end of the decade. While other benchmarks from Google DeepMind and ARC Prize Foundation indicate that Artificial General Intelligence (AGI) is still far off, the 2026 International AI Safety Report noted improvements in general-purpose AI but also inconsistencies, with AI translation remaining a highly important area.
Key Takeaways
- From 2015 to 2022, Translated's AI reduced the time required for human editors to correct machine-translated text from 3.5 seconds per word to 2 seconds per word.
- Translated predicts machine translation could achieve human-level editing effort by 2030, where editing AI output is as efficient as editing human work.
- Google DeepMind's 'Humanity's Last Exam' benchmark shows frontier models like Gemini 3 Pro (38.3%) and GPT-5 (25.3%) still have significant gaps compared to human performance.
- The ARC Prize Foundation's ARC-AGI-3 benchmark, testing interactive reasoning, found present AI systems score under 1%, while humans achieve 100%.
- The 2026 International AI Safety Report noted AI models show inconsistencies, excelling in complex tasks yet struggling with simpler ones.
Why It Matters
Translated's data suggests AI could soon match human efficiency in translation editing, impacting localization workflows and content scalability for global streaming platforms. While specific metrics point to rapid progress in narrow AI tasks like translation, broader AGI benchmarks indicate a substantial gap in general intelligence. Companies should monitor the development of AI-driven tools that reduce post-editing time, but remain aware of the continued necessity for human oversight in complex cognitive tasks.
Additional Context
Translated CEO Marco Trombetti, in a December 2022 company post, quantified the speed toward AI singularity using machine translation trends. He noted that if current progress continues, professional translators could spend the same amount of time correcting AI translations as they do peer translations within several years (per Translated, Dec 2022). This 'singularity in translation' could lead to a tenfold increase in demand for professional translation services and 100 times growth in machine translation demand, as observed by Translated from clients increasing content localization (per Translated, Dec 2022).
Translated emphasizes a 'perfect symbiosis' between humans and machines, where AI acts as a valuable tool for professionals, not a replacement. Their data indicates that translators using their Matecat tool have seen average earnings increase by 25%, with some multiplying revenues three or four times (per Translated, Dec 2022). This highlights a shift to a 'machine-first, human-optimized' approach in large-scale translation, integrating expert human feedback to continuously improve machine translation output quality (per Imminent, Translated's Research Center, undated). The company's ModernMT technology, designed for adaptive learning, is critical to this ongoing improvement, providing a substantial advantage over generic public MT engines (per Imminent, Translated's Research Center, undated). Despite advancements, challenges remain in handling idiomatic expressions, complex grammar, and real-time processing, necessitating robust frameworks for fair and secure usage (per Translated, undated).
Read full article at msn.com
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source